检索结果-内蒙古大学图书馆

An Adversarial Feature Distillation Method for Audio Classification

IEEE ACCESS 2019年 7卷 105319-105330页

作者： Gao, Liang Mi, Haibo Zhu, Boqing Feng, Dawei Li, Yicong Peng, Yuxing Natl Univ Def Technol Natl Key Lab Parallel & Distributed Proc Changsha 410073 Hunan Peoples R China

The audio classification task aims to discriminate between different audio signal types. In this task, deep neural networks have achieved better performance than the traditional shallow architecture-based machine-learning method. However, deep neural networks often require huge computational and storage requirements that hinder the deployment in embedded devices. In this paper, we proposed a distillation method which transfers knowledge from well-trained networks to a small network, and the method can compress model size while improving audio classification precision. The contributions of the proposed method are two folds: a multi-level feature distillation method was proposed and an adversarial learning strategy was employed to improve the knowledge transfer. The extensive experiments are conducted on three audio classification tasks, audio scene classification, general audio tagging, and speech command recognition. The experimental results demonstrate that: the small network can provide better performance while achieves the calculated amount of floating-point operations per second (FLOPS) compression ratio of 76:1 and parameters compression ratio of 3:1.

关键词： Convolutional neural networks audio tagging knowledge distillation model compression

来源：评论

学校读者我要写书评

暂无评论

Research and performance evaluation of data replication technology in distributed storage systems

引用

COMPUTERS & MATHEMATICS WITH APPLICATIONS 2006年第11期51卷 1625-1632页

作者： Wang, Yijie Li, Sijun Natl Univ Def Technol Inst Comp Natl Lab Parallel & Distributed Proc Changsha 410073 Peoples R China

Replication is a key technology of distributed storage systems. In this paper, an indirect replication algorithm is proposed following the intrinsic characteristic of distributed storage systems and the peer-to-peer model. In the indirect replication algorithm, the data object is partitioned into several data blocks. These data blocks are encoded in order that there is data redundancy between data blocks. Comparing with the traditional replication algorithm, the indirect replication algorithm has less granularity of replication, less bandwidth and storage costs, and provides higher availability, durability, and security. The performance evaluation shows that the encoding and decoding times are proportional to the data size, and that the irregular cascade bipartite graphs are of great advantage in improving the success ratio of data recovery. Finally, if the number of data blocks used to recover the data object is larger than a certain value, the success ratio of data recovery approaches 100%. (c) 2006 Elsevier Ltd. All rights reserved.

关键词： distributed storage system replication data availability code performance evaluation

来源：评论

学校读者我要写书评

暂无评论

Exploring frame segmentation networks for temporal action localization

引用

JOURNAL OF VISUAL COMMUNICATION AND IMAGE REPRESENTATION 2019年第May期61卷 296-302页

作者： Yang, Ke Shen, Xiaolong Qiao, Peng Li, Shijie Li, Dongsheng Dou, Yong Natl Univ Def Technol Coll Comp Natl Lab Parallel & Distributed Proc Changsha Hunan Peoples R China

Temporal action localization is an important task of computer vision. Though many methods have been proposed, it still remains an open question how to predict the temporal location of action segments precisely. Most state-of-the-art works train action classifiers on video segments pre-determined by action proposal. However, recent work found that a desirable model should move beyond segment-level and make dense predictions at a fine granularity in time to determine precise temporal boundaries. In this paper, we propose a Frame Segmentation Network (FSN) that places a temporal CNN on top of the 2D spatial CNNs. Spatial CNNs are responsible for abstracting semantics in spatial dimension while temporal CNN is responsible for introducing temporal context information and performing dense predictions. The proposed FSN can make dense predictions at frame-level for a video clip using both spatial and temporal context information. FSN is trained in an end-to-end manner, so the model can be optimized in spatial and temporal domain jointly. We also adapt FSN to use it in weakly supervised scenario (WFSN), where only video level labels are provided when training. Experiment results on public dataset show that FSN achieves superior performance in both frame-level action localization and temporal action localization. (C) 2019 Elsevier Inc. All rights reserved.

关键词： Action detection Temporal action localization Convolutional Neural Network

来源：评论

学校读者我要写书评

暂无评论

An automatic learning rate decay strategy for stochastic gradient descent optimization methods in neural networks

引用

INTERNATIONAL JOURNAL OF INTELLIGENT SYSTEMS 2022年第10期37卷 7334-7355页

作者： Wang, Kang Dou, Yong Sun, Tao Qiao, Peng Wen, Dong Natl Univ Def Technol Sch Comp Natl Lab Parallel & Distributed Proc Changsha 410073 Peoples R China

Stochastic Gradient Descent (SGD) series optimization methods play the vital role in training neural networks, attracting growing attention in science and engineering fields of the intelligent system. The choice of learning rates affects the convergence rate of SGD series optimization methods. Currently, learning rate adjustment strategies mainly face the following problems: (1) The traditional learning rate decay method mainly adopts manual manner during training iterations, the small learning rate produced from which causes slow convergence in training neural networks. (2) Adaptive method (e.g., Adam) has poor generalization performance. To alleviate the above issues, we propose a novel automatic learning rate decay strategy for SGD optimization methods in neural networks. On the basis of the observation that the convergence rate's upper bound enjoys minimization in a specific iteration concerning the current learning rate, we first present the expression of the current learning rate determined by historical learning rates. And merely one extra parameter is initialized to generate automatic decreasing learning rates during the training process. Our proposed approach is applied to SGD and Momentum SGD optimization algorithms, and concrete theoretical proof explains its convergence. Numerical simulations are conducted on the MNIST and Cifar-10 data sets with different neural networks. Experimental results show that our algorithm outperforms existing classical ones, achieving faster convergence rate, better stability, and generalization performance in neural network training. It also lays a foundation for large-scale parallel search of initial parameters in intelligent systems.

关键词： automatic learning rate decay strategy fast convergence generalization performance neural networks stochastic gradient descent optimization algorithms training stability

来源：评论

学校读者我要写书评

暂无评论

ESDU: An elastic stripe-based delta update method for erasure-coded cross-data center storage systems

引用

JOURNAL OF parallel AND distributed COMPUTING 2022年第0期167卷 173-186页

作者： Bao, Han Wang, Yijie Natl Univ Def Technol Sch Comp Natl Lab Parallel & Distributed Proc Changsha 410073 Peoples R China

The erasure-coded cross-data center storage system can achieve high disaster tolerance and low redundancy. But as it has large cross-data center update traffic, its data update time is long. In erasure-coded storage systems, each data object is sequentially divided into several stripes, and each stripe consists of several data packets. When erasure-coded stripes do not undergo insertion or deletion, existing work can effectively reduce cross-data center update traffic by performing delta updates- delta update methods can update the old stripe without transferring matched packets that are new and old stripes' duplicate packets with the same offset-within-stripe. However, because existing delta update methods' stripe size is fixed, when a stripe undergoes insertion or deletion, its subsequent stripes' duplicate packets' offset-within-stripe will change. In this scenery, the matched packet number is small, resulting in large cross-data center update traffic. This paper proposes an elastic stripe-based delta update method for erasure-coded cross-data center storage systems (ESDU). Under insertion or deletion, ESDS tries to avoid duplicate packets' offset-within-stripe changing (i.e., maximizing matched packets) by adjusting the stripes' size flexibly according to the duplicate packet locating result. So, it can reduce cross-data center traffic. Moreover, ESDU can optimize stripes' update topology based on the location information of storage nodes to reduce cross-data center update traffic further. In addition, we implement an erasure-coded cross-data center storage system adopting ESDU, called ECESD. Experiments with the workloads derived from EduCoder's real-world trace show that compared with the existing erasure-coded cross-data center storage system adopting the fixed stripe-based delta update method, ECESD reduces average update time by 89.6%. Moreover, compared with a replication-based storage system with a delta update method (HadoopRsync), ECESD achieves an 8.3%

关键词： Erasure code Disaster tolerance distributed storage system Delta update method Cross-data center storage

来源：评论

学校读者我要写书评

暂无评论

A Novel Architecture to Eliminate Bottlenecks in a parallel Tiled QRD Algorithm for Future MIMO Systems

引用

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS II-EXPRESS BRIEFS 2017年第1期64卷 26-30页

作者： Liu, Cang Xing, Zuocheng Yuan, Luechao Tang, Chuan Zhang, Yang Natl Univ Def Technol Natl Lab Parallel & Distributed Proc Changsha 410073 Hunan Peoples R China

QR decomposition (QRD) is one of the performance bottlenecks in a lot of high performance wireless communication algorithms and should have the flexibility property for future multiple-input multiple-output systems. However, the existing QRD architectures only focus on several fixed dimension matrices. The parallel tiled QRD algorithm is a perfect choice to implement QRD for its flexibility and modularity property. The size of the tile is set to 2 x 2 instead of the traditional 200 x 200 or more to support flexible antenna configurations in this brief. Using a look-ahead technique and the property of unitary matrix, a novel algorithm based on a modified Gram-Schmidt (MGS) algorithm is proposed for the bottleneck operations (GEQRT and TTQRT) of the parallel tiled QRD algorithm. A corresponding hardware architecture is also designed with the proposed algorithm. The implementation results show that the hardware architecture based on the proposed algorithm achieves a 2.7x reduction in normalized processing latency, compared with the one based on the traditional MGS algorithm.

关键词： Modified Gram-Schmidt (MGS) parallel tiled QR decomposition (QRD) wireless communication

来源：评论

学校读者我要写书评

暂无评论

High Precision Low Complexity Matrix Inversion Based on Newton Iteration for Data Detection in the Massive MIMO

引用

IEEE COMMUNICATIONS LETTERS 2016年第3期20卷 490-493页

作者： Tang, Chuan Liu, Cang Yuan, Luechao Xing, Zuocheng Natl Univ Def Technol Natl Lab Parallel & Distributed Proc Changsha 410073 Hunan Peoples R China

Currently, massive multiple-input multiple-output (MIMO) is one of the most promising wireless transmission technologies for 5G. Massive MIMO requires handling with large-scale matrix computation, especially for matrix inversion. In this letter, we find that matrix inversion based on Newton iteration (NI) is suitable for data detection in massive MIMO system. In contrast with recently proposed polynomial expansion (PE) method for matrix inversion, we analyze both the algorithm complexity and precision in detail, and propose a diagonal band Newton iteration (DBNI) method, which is an approximate method for NI. Compared with PE method, DBNI can obtain higher precision and approximately equal complexity, and we give an explanation of how to select the bandwidth of DBNI.

关键词： Massive MIMO data detection matrix inversion Newton iteration Neumann series

来源：评论

学校读者我要写书评

暂无评论

A simple coverage-evaluating approach for wireless sensor networks with arbitrary sensing areas

引用

INFORMATION procESSING LETTERS 2008年第4期106卷 159-161页

作者： Fan, GaoJun Jin, ShiYao Natl Univ Def Technol Natl Lab Parallel & Distributed Proc Changsha 410073 Hunan Peoples R China

We present a simple and fast deterministic solution to the area coverage problem in wireless sensor networks. The task is to determine whether all points in a region are covered by a given set of sensors, where each sensor may have any arbitrary sensing shape. We transform the area coverage problem to the intersection points' coverage problem, which is simpler and more suitable for evaluating the area coverage problem than previous approaches in our view. (c) 2007 Elsevier B.V. All rights reserved.

关键词： sensor network computational geometry k-coverage arbitrary sensing areas

来源：评论

学校读者我要写书评

暂无评论

Fine-grained parallel RNAalifold algorithm for RNA secondary structure prediction on FPGA

引用

BMC BIOINFORMATICS 2009年第s1期10卷 1-14页

作者： Xia, Fei Dou, Yong Zhou, Xingming Yang, Xuejun Xu, Jiaqing Zhang, Yang Natl Univ Def Technol Dept Comp Sci Natl Lab Parallel & Distributed Proc Changsha 410073 Peoples R China

Background: In the field of RNA secondary structure prediction, the RNAalifold algorithm is one of the most popular methods using free energy minimization. However, general-purpose computers including parallel computers or multi-core computers exhibit parallel efficiency of no more than 50%. Field Programmable Gate-Array (FPGA) chips provide a new approach to accelerate RNAalifold by exploiting fine-grained custom design. Results: RNAalifold shows complicated data dependences, in which the dependence distance is variable, and the dependence direction is also across two dimensions. We propose a systolic array structure including one master processing Element (PE) and multiple slave PEs for fine grain hardware implementation on FPGA. We exploit data reuse schemes to reduce the need to load energy matrices from external memory. We also propose several methods to reduce energy table parameter size by 80%. Conclusion: To our knowledge, our implementation with 16 PEs is the only FPGA accelerator implementing the complete RNAalifold algorithm. The experimental results show a factor of 12.2 speedup over the RNAalifold (ViennaPackage - 1.6.5) software for a group of aligned RNA sequences with 2981-residue running on a Personal Computer (PC) platform with Pentium 4 2.6 GHz CPU.

关键词： Free Energy processing Element Energy Matrice FPGA Chip processing Element Array

来源：评论

学校读者我要写书评

暂无评论

Simulation Study of Large-Scale Charge Sharing Mitigation Using Seamless Guard Band

引用

IEEE TRANSACTIONS ON DEVICE AND MATERIALS RELIABILITY 2017年第1期17卷 176-183页

作者： Huang, Pengcheng Chen, Shuming Chen, Jianjun Bin, Liang Wu, Zhenyu Natl Univ Def Technol Natl Lab Parallel & Distributed Proc Coll Comp Changsha 410073 Hunan Peoples R China

In this paper, a novel seamless guard band (SGB) technique for charge sharing mitigation is studied using 3-D TCAD numerical simulations. The simulations results in 65-nm twin-well bulk CMOS technology indicate that the SGB technique can not only mitigate the single-event transient pulsewidth greatly but also mitigate the charge sharing between logical nodes or logical cells significantly. The simulation results also indicate that the SGB technique is superior to the conventional guard band (GB) technique, for it is more beneficial for parasitic bipolar effect mitigation. Using SGB technique, the single-event double-transient (SEDT) generation is mitigated completely under low LET particle (LET <= 40 MeV . cm(2)/mg) radiation, and the SEDT pulsewidth is mitigated > 50% even with the LET of 80 MeV . cm(2)/mg, which is > 25% from GB technique. Finally, the SGB technique can be applied to the construction of a radiation-hardened standard cell library conveniently, and its area penalty is 1-1.67 x, which is the same with that of the GB technique.

关键词： Seamless guard band (SGB) single event double transient (SEDT) single-event transient (SET) charge sharing parasitic bipolar effect (PBE)

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：